NZ Police cut roughly two dozen jobs in its file management unit in October, part of a restructure that disestablished, created or shifted more than 60 positions. The people who had spent two years fearing an AI-driven cull, Transcription Services, were spared. For now. They have been handed a separate “modernisation programme” instead.
That is the real story for business. AI pressure is not landing first on the glamorous jobs everyone predicted. It is landing on repetitive, measurable administrative work, and it is arriving tangled up with budget cuts in a way that makes it almost impossible to say cleanly what caused what.
A denial that sits awkwardly with the paperwork
Sarah Hutchings, Police’s national manager for front-line file management and transcription services, told RNZ that “Artificial Intelligence was not a factor of workforce reductions in this review.” Yet Police’s own July 2026 restructure document states: “As these technologies become fully embedded, they will directly influence the size, structure, and skill mix of our workforce.”
Both statements can be technically true. But the history makes the distinction look thin. After a six-month 2025 trial of Whisper AI was undermined by officers breaching safeguards, Police banned it from transcribing English, te reo Maori and Pacific languages. Internal documents described that ban as “not driven by concerns about technological risk, but rather by the potential impact on specific roles”. Since then Police has added Meta’s translation tool, European AI for child exploitation cases and, this week, an Axon tool to transcribe interviews. The PSA says it cannot establish that staff were properly consulted before any of it.
When an organisation pauses a tool specifically to protect jobs, then reintroduces it while restructuring the adjacent team, staff will join the dots whatever the press line says.
The back office was already running on fumes
This restructure did not arrive in a vacuum. Budget 2024 asked Police to find $55 million in savings, initially through around 175 roles. Across the wider public service, an estimated 1,723 redundancies were made in the year to September 2024, double the previous year.
Then an April 2026 review found the police back office was a “critical enabler” that had been “underpowered and undervalued”, with ministers warned that Police were not sure how to make further savings safely. Cutting it again, with automation in the background, is a bet that technology will fill the gap. That bet may well be right. But it needs evidence.
Measurement is the part almost nobody does
Here Police deserves genuine credit. A BusinessDesk investigation in April 2026 found Police was the only agency with real before-and-after data: its $8.2 million Appian file assessment platform cut processing from 8-10 minutes to 1-3 minutes per case, saving 18,302 hours and $996,769 a year. Elsewhere the picture was dire. MBIE admitted its productivity gains were “self-reported and not empirically tested”, and IRD found no enterprise-wide framework to measure gains at all.
That is a payback period of around eight years on salary savings alone, which is respectable for core infrastructure. It is also exactly the kind of routine, rules-based processing that file management involves. The tasks went first. The roles followed.
Private-sector experience points the same way. In 2025, McKinsey estimated 60-70% of activities in office support, customer service and professional services could be automated. A University of Auckland analysis in April 2026 found AI was mostly automating tasks rather than whole roles, but that more than half of NZ employers surveyed reported significant displacement and slower entry-level hiring.
Upskill first, restructure second
Not everyone thinks headcount should be the scoreboard. Dr Viveca Pavon-Harr, Accenture’s global AI and data lead for public sector, challenges anyone to “show me a public government office where you have an overflow of employees, and everything’s already up to date”, and BusinessDesk’s reporting concludes that “a massive upskilling push really needs to precede any AI-driven transformation”.
That is sound commercial advice, not union sentiment. Backlogs, compliance work and slow service are where most back offices actually hurt. Freeing capacity to clear them can be worth more than the redundancy saving.
What employers should take from this
The Police case is a free lesson in how not to communicate automation. Measure gains properly before and after, as Police did with Appian. Consult before tools go live, not after. And do not insist AI played no part in a restructure when your own planning documents say it will shape the workforce. Staff can read.
Automation is coming for invoicing, records, data entry, claims processing and scheduling first, because that work is repetitive and easy to count. Firms that are honest about it, and redeploy people into higher-value work, will keep the trust they need for the next round. Transcription Services is already in line for it.
Sources
- RNZ: AI ‘panic’: Back office jobs cut – but not the expected roles (2026-11-04)
- RNZ: ‘Underpowered’: The police back-office holding back the front line (2026-06-30)
- FYI.org.nz: OIA Response IR 01 24 36856 – Budget 2024 Savings Proposal (2024-11-06)
- Public Service Commission: Public Service Workforce Data – September Quarterly Update (2024-12-03)
- BusinessDesk: ‘Slippery slope’: Concerns millions spent on AI may not be delivering for public (2026-04-07)
- The Conversation: AI is peeling back the layers of ‘low-value’ work, NZ may be well placed to adapt (2025-08-12)
- University of Auckland: If entry-level jobs are disappearing, how do young people get started? (2026-04-21)
- BusinessDesk: Government’s AI plan for public sector cuts questioned by tech leaders (2026-09-22)
Join the discussion
Add useful context, ask a good question, or challenge an idea — keep it specific and respectful.